* of the Software.
*/
-#define _FILE_OFFSET_BITS 64
#include "radeon_drm_cs.h"
#include "util/u_hash_table.h"
#include "util/u_double_list.h"
#include "os/os_thread.h"
#include "os/os_mman.h"
+#include "os/os_time.h"
#include "state_tracker/drm_driver.h"
#include <sys/ioctl.h>
#include <xf86drm.h>
#include <errno.h>
-
-/*
- * this are copy from radeon_drm, once an updated libdrm is released
- * we should bump configure.ac requirement for it and remove the following
- * field
- */
-#define RADEON_BO_FLAGS_MACRO_TILE 1
-#define RADEON_BO_FLAGS_MICRO_TILE 2
-#define RADEON_BO_FLAGS_MICRO_TILE_SQUARE 0x20
-
-#ifndef DRM_RADEON_GEM_WAIT
-#define DRM_RADEON_GEM_WAIT 0x2b
-
-#define RADEON_GEM_NO_WAIT 0x1
-#define RADEON_GEM_USAGE_READ 0x2
-#define RADEON_GEM_USAGE_WRITE 0x4
-
-struct drm_radeon_gem_wait {
- uint32_t handle;
- uint32_t flags; /* one of RADEON_GEM_* */
-};
-
-#endif
-
-#ifndef RADEON_VA_MAP
-
-#define RADEON_VA_MAP 1
-#define RADEON_VA_UNMAP 2
-
-#define RADEON_VA_RESULT_OK 0
-#define RADEON_VA_RESULT_ERROR 1
-#define RADEON_VA_RESULT_VA_EXIST 2
-
-#define RADEON_VM_PAGE_VALID (1 << 0)
-#define RADEON_VM_PAGE_READABLE (1 << 1)
-#define RADEON_VM_PAGE_WRITEABLE (1 << 2)
-#define RADEON_VM_PAGE_SYSTEM (1 << 3)
-#define RADEON_VM_PAGE_SNOOPED (1 << 4)
-
-struct drm_radeon_gem_va {
- uint32_t handle;
- uint32_t operation;
- uint32_t vm_id;
- uint32_t flags;
- uint64_t offset;
-};
-
-#define DRM_RADEON_GEM_VA 0x2b
-#endif
-
-
+#include <fcntl.h>
+#include <stdio.h>
extern const struct pb_vtbl radeon_bo_vtbl;
-
static INLINE struct radeon_bo *radeon_bo(struct pb_buffer *bo)
{
assert(bo->vtbl == &radeon_bo_vtbl);
/* Winsys. */
struct radeon_drm_winsys *rws;
- /* List of buffer handles and its mutex. */
+ /* List of buffer GEM names. Protected by bo_handles_mutex. */
+ struct util_hash_table *bo_names;
+ /* List of buffer handles. Protectded by bo_handles_mutex. */
struct util_hash_table *bo_handles;
+ /* List of buffer virtual memory ranges. Protectded by bo_handles_mutex. */
+ struct util_hash_table *bo_vas;
pipe_mutex bo_handles_mutex;
pipe_mutex bo_va_mutex;
static void radeon_bo_wait(struct pb_buffer *_buf, enum radeon_bo_usage usage)
{
struct radeon_bo *bo = get_radeon_bo(_buf);
+ struct drm_radeon_gem_wait_idle args = {0};
while (p_atomic_read(&bo->num_active_ioctls)) {
sched_yield();
}
- /* XXX use this when it's ready */
- /*if (bo->rws->info.drm_minor >= 12) {
- struct drm_radeon_gem_wait args = {};
- args.handle = bo->handle;
- args.flags = usage;
- while (drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT,
- &args, sizeof(args)) == -EBUSY);
- } else*/ {
- struct drm_radeon_gem_wait_idle args;
- memset(&args, 0, sizeof(args));
- args.handle = bo->handle;
- while (drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT_IDLE,
- &args, sizeof(args)) == -EBUSY);
- }
+ args.handle = bo->handle;
+ while (drmCommandWrite(bo->rws->fd, DRM_RADEON_GEM_WAIT_IDLE,
+ &args, sizeof(args)) == -EBUSY);
}
static boolean radeon_bo_is_busy(struct pb_buffer *_buf,
enum radeon_bo_usage usage)
{
struct radeon_bo *bo = get_radeon_bo(_buf);
+ struct drm_radeon_gem_busy args = {0};
if (p_atomic_read(&bo->num_active_ioctls)) {
return TRUE;
}
- /* XXX use this when it's ready */
- /*if (bo->rws->info.drm_minor >= 12) {
- struct drm_radeon_gem_wait args = {};
- args.handle = bo->handle;
- args.flags = usage | RADEON_GEM_NO_WAIT;
- return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_WAIT,
- &args, sizeof(args)) != 0;
- } else*/ {
- struct drm_radeon_gem_busy args;
- memset(&args, 0, sizeof(args));
- args.handle = bo->handle;
- return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_BUSY,
- &args, sizeof(args)) != 0;
- }
+ args.handle = bo->handle;
+ return drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_BUSY,
+ &args, sizeof(args)) != 0;
+}
+
+static enum radeon_bo_domain get_valid_domain(enum radeon_bo_domain domain)
+{
+ /* Zero domains the driver doesn't understand. */
+ domain &= RADEON_DOMAIN_VRAM_GTT;
+
+ /* If no domain is set, we must set something... */
+ if (!domain)
+ domain = RADEON_DOMAIN_VRAM_GTT;
+
+ return domain;
+}
+
+static enum radeon_bo_domain radeon_bo_get_initial_domain(
+ struct radeon_winsys_cs_handle *buf)
+{
+ struct radeon_bo *bo = (struct radeon_bo*)buf;
+ struct drm_radeon_gem_op args;
+
+ if (bo->rws->info.drm_minor < 38)
+ return RADEON_DOMAIN_VRAM_GTT;
+
+ memset(&args, 0, sizeof(args));
+ args.handle = bo->handle;
+ args.op = RADEON_GEM_OP_GET_INITIAL_DOMAIN;
+
+ drmCommandWriteRead(bo->rws->fd, DRM_RADEON_GEM_OP,
+ &args, sizeof(args));
+
+ /* GEM domains and winsys domains are defined the same. */
+ return get_valid_domain(args.value);
}
static uint64_t radeon_bomgr_find_va(struct radeon_bomgr *mgr, uint64_t size, uint64_t alignment)
struct radeon_bo_va_hole *hole, *n;
uint64_t offset = 0, waste = 0;
+ alignment = MAX2(alignment, 4096);
+ size = align(size, 4096);
+
pipe_mutex_lock(mgr->bo_va_mutex);
/* first look for a hole */
LIST_FOR_EACH_ENTRY_SAFE(hole, n, &mgr->va_holes, list) {
offset = hole->offset;
- waste = 0;
- if (alignment) {
- waste = offset % alignment;
- waste = waste ? alignment - waste : 0;
- }
+ waste = offset % alignment;
+ waste = waste ? alignment - waste : 0;
offset += waste;
if (offset >= (hole->offset + hole->size)) {
continue;
}
offset = mgr->va_offset;
- waste = 0;
- if (alignment) {
- waste = offset % alignment;
- waste = waste ? alignment - waste : 0;
- }
+ waste = offset % alignment;
+ waste = waste ? alignment - waste : 0;
if (waste) {
n = CALLOC_STRUCT(radeon_bo_va_hole);
n->size = waste;
return offset;
}
-static void radeon_bomgr_force_va(struct radeon_bomgr *mgr, uint64_t va, uint64_t size)
-{
- pipe_mutex_lock(mgr->bo_va_mutex);
- if (va >= mgr->va_offset) {
- if (va > mgr->va_offset) {
- struct radeon_bo_va_hole *hole;
- hole = CALLOC_STRUCT(radeon_bo_va_hole);
- if (hole) {
- hole->size = va - mgr->va_offset;
- hole->offset = mgr->va_offset;
- list_add(&hole->list, &mgr->va_holes);
- }
- }
- mgr->va_offset = va + size;
- } else {
- struct radeon_bo_va_hole *hole, *n;
- uint64_t hole_end, va_end;
-
- /* Prune/free all holes that fall into the range
- */
- LIST_FOR_EACH_ENTRY_SAFE(hole, n, &mgr->va_holes, list) {
- hole_end = hole->offset + hole->size;
- va_end = va + size;
- if (hole->offset >= va_end || hole_end <= va)
- continue;
- if (hole->offset >= va && hole_end <= va_end) {
- list_del(&hole->list);
- FREE(hole);
- continue;
- }
- if (hole->offset >= va)
- hole->offset = va_end;
- else
- hole_end = va;
- hole->size = hole_end - hole->offset;
- }
- }
- pipe_mutex_unlock(mgr->bo_va_mutex);
-}
-
static void radeon_bomgr_free_va(struct radeon_bomgr *mgr, uint64_t va, uint64_t size)
{
struct radeon_bo_va_hole *hole;
+ size = align(size, 4096);
+
pipe_mutex_lock(mgr->bo_va_mutex);
if ((va + size) == mgr->va_offset) {
mgr->va_offset = va;
memset(&args, 0, sizeof(args));
- if (bo->name) {
- pipe_mutex_lock(bo->mgr->bo_handles_mutex);
- util_hash_table_remove(bo->mgr->bo_handles,
- (void*)(uintptr_t)bo->name);
- pipe_mutex_unlock(bo->mgr->bo_handles_mutex);
+ pipe_mutex_lock(bo->mgr->bo_handles_mutex);
+ util_hash_table_remove(bo->mgr->bo_handles, (void*)(uintptr_t)bo->handle);
+ if (bo->flink_name) {
+ util_hash_table_remove(bo->mgr->bo_names,
+ (void*)(uintptr_t)bo->flink_name);
}
+ pipe_mutex_unlock(bo->mgr->bo_handles_mutex);
if (bo->ptr)
os_munmap(bo->ptr, bo->base.size);
drmIoctl(bo->rws->fd, DRM_IOCTL_GEM_CLOSE, &args);
if (mgr->va) {
- radeon_bomgr_free_va(mgr, bo->va, bo->va_size);
+ radeon_bomgr_free_va(mgr, bo->va, bo->base.size);
}
pipe_mutex_destroy(bo->map_mutex);
+
+ if (bo->initial_domain & RADEON_DOMAIN_VRAM)
+ bo->rws->allocated_vram -= align(bo->base.size, 4096);
+ else if (bo->initial_domain & RADEON_DOMAIN_GTT)
+ bo->rws->allocated_gtt -= align(bo->base.size, 4096);
FREE(bo);
}
+void *radeon_bo_do_map(struct radeon_bo *bo)
+{
+ struct drm_radeon_gem_mmap args = {0};
+ void *ptr;
+
+ /* Return the pointer if it's already mapped. */
+ if (bo->ptr)
+ return bo->ptr;
+
+ /* Map the buffer. */
+ pipe_mutex_lock(bo->map_mutex);
+ /* Return the pointer if it's already mapped (in case of a race). */
+ if (bo->ptr) {
+ pipe_mutex_unlock(bo->map_mutex);
+ return bo->ptr;
+ }
+ args.handle = bo->handle;
+ args.offset = 0;
+ args.size = (uint64_t)bo->base.size;
+ if (drmCommandWriteRead(bo->rws->fd,
+ DRM_RADEON_GEM_MMAP,
+ &args,
+ sizeof(args))) {
+ pipe_mutex_unlock(bo->map_mutex);
+ fprintf(stderr, "radeon: gem_mmap failed: %p 0x%08X\n",
+ bo, bo->handle);
+ return NULL;
+ }
+
+ ptr = os_mmap(0, args.size, PROT_READ|PROT_WRITE, MAP_SHARED,
+ bo->rws->fd, args.addr_ptr);
+ if (ptr == MAP_FAILED) {
+ pipe_mutex_unlock(bo->map_mutex);
+ fprintf(stderr, "radeon: mmap failed, errno: %i\n", errno);
+ return NULL;
+ }
+ bo->ptr = ptr;
+ pipe_mutex_unlock(bo->map_mutex);
+
+ return bo->ptr;
+}
+
static void *radeon_bo_map(struct radeon_winsys_cs_handle *buf,
struct radeon_winsys_cs *rcs,
enum pipe_transfer_usage usage)
{
struct radeon_bo *bo = (struct radeon_bo*)buf;
struct radeon_drm_cs *cs = (struct radeon_drm_cs*)rcs;
- struct drm_radeon_gem_mmap args = {0};
- void *ptr;
/* If it's not unsynchronized bo_map, flush CS if needed and then wait. */
if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
* (neither one is changing it).
*
* Only check whether the buffer is being used for write. */
- if (radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
- cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC);
+ if (cs && radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
+ cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC, NULL);
return NULL;
}
return NULL;
}
} else {
- if (radeon_bo_is_referenced_by_cs(cs, bo)) {
- cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC);
+ if (cs && radeon_bo_is_referenced_by_cs(cs, bo)) {
+ cs->flush_cs(cs->flush_data, RADEON_FLUSH_ASYNC, NULL);
return NULL;
}
}
}
} else {
+ uint64_t time = os_time_get_nano();
+
if (!(usage & PIPE_TRANSFER_WRITE)) {
/* Mapping for read.
*
* (neither one is changing it).
*
* Only check whether the buffer is being used for write. */
- if (radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
- cs->flush_cs(cs->flush_data, 0);
+ if (cs && radeon_bo_is_referenced_by_cs_for_write(cs, bo)) {
+ cs->flush_cs(cs->flush_data, 0, NULL);
}
radeon_bo_wait((struct pb_buffer*)bo,
RADEON_USAGE_WRITE);
} else {
/* Mapping for write. */
- if (radeon_bo_is_referenced_by_cs(cs, bo)) {
- cs->flush_cs(cs->flush_data, 0);
- } else {
- /* Try to avoid busy-waiting in radeon_bo_wait. */
- if (p_atomic_read(&bo->num_active_ioctls))
- radeon_drm_cs_sync_flush(cs);
+ if (cs) {
+ if (radeon_bo_is_referenced_by_cs(cs, bo)) {
+ cs->flush_cs(cs->flush_data, 0, NULL);
+ } else {
+ /* Try to avoid busy-waiting in radeon_bo_wait. */
+ if (p_atomic_read(&bo->num_active_ioctls))
+ radeon_drm_cs_sync_flush(rcs);
+ }
}
radeon_bo_wait((struct pb_buffer*)bo, RADEON_USAGE_READWRITE);
}
- }
- }
-
- /* Return the pointer if it's already mapped. */
- if (bo->ptr)
- return bo->ptr;
-
- /* Map the buffer. */
- pipe_mutex_lock(bo->map_mutex);
- /* Return the pointer if it's already mapped (in case of a race). */
- if (bo->ptr) {
- pipe_mutex_unlock(bo->map_mutex);
- return bo->ptr;
- }
- args.handle = bo->handle;
- args.offset = 0;
- args.size = (uint64_t)bo->base.size;
- if (drmCommandWriteRead(bo->rws->fd,
- DRM_RADEON_GEM_MMAP,
- &args,
- sizeof(args))) {
- pipe_mutex_unlock(bo->map_mutex);
- fprintf(stderr, "radeon: gem_mmap failed: %p 0x%08X\n",
- bo, bo->handle);
- return NULL;
- }
- ptr = os_mmap(0, args.size, PROT_READ|PROT_WRITE, MAP_SHARED,
- bo->rws->fd, args.addr_ptr);
- if (ptr == MAP_FAILED) {
- pipe_mutex_unlock(bo->map_mutex);
- fprintf(stderr, "radeon: mmap failed, errno: %i\n", errno);
- return NULL;
+ bo->mgr->rws->buffer_wait_time += os_time_get_nano() - time;
+ }
}
- bo->ptr = ptr;
- pipe_mutex_unlock(bo->map_mutex);
- return bo->ptr;
+ return radeon_bo_do_map(bo);
}
static void radeon_bo_unmap(struct radeon_winsys_cs_handle *_buf)
radeon_bo_get_base_buffer,
};
+#ifndef RADEON_GEM_GTT_WC
+#define RADEON_GEM_GTT_WC (1 << 2)
+#endif
+#ifndef RADEON_GEM_CPU_ACCESS
+/* BO is expected to be accessed by the CPU */
+#define RADEON_GEM_CPU_ACCESS (1 << 3)
+#endif
+#ifndef RADEON_GEM_NO_CPU_ACCESS
+/* CPU access is not expected to work for this BO */
+#define RADEON_GEM_NO_CPU_ACCESS (1 << 4)
+#endif
+
static struct pb_buffer *radeon_bomgr_create_bo(struct pb_manager *_mgr,
pb_size size,
const struct pb_desc *desc)
args.size = size;
args.alignment = desc->alignment;
args.initial_domain = rdesc->initial_domains;
+ args.flags = 0;
+
+ if (rdesc->flags & RADEON_FLAG_GTT_WC)
+ args.flags |= RADEON_GEM_GTT_WC;
+ if (rdesc->flags & RADEON_FLAG_CPU_ACCESS)
+ args.flags |= RADEON_GEM_CPU_ACCESS;
+ if (rdesc->flags & RADEON_FLAG_NO_CPU_ACCESS)
+ args.flags |= RADEON_GEM_NO_CPU_ACCESS;
if (drmCommandWriteRead(rws->fd, DRM_RADEON_GEM_CREATE,
&args, sizeof(args))) {
fprintf(stderr, "radeon: size : %d bytes\n", size);
fprintf(stderr, "radeon: alignment : %d bytes\n", desc->alignment);
fprintf(stderr, "radeon: domains : %d\n", args.initial_domain);
+ fprintf(stderr, "radeon: flags : %d\n", args.flags);
return NULL;
}
bo->rws = mgr->rws;
bo->handle = args.handle;
bo->va = 0;
+ bo->initial_domain = rdesc->initial_domains;
pipe_mutex_init(bo->map_mutex);
if (mgr->va) {
struct drm_radeon_gem_va va;
- bo->va_size = align(size, 4096);
- bo->va = radeon_bomgr_find_va(mgr, bo->va_size, desc->alignment);
+ bo->va = radeon_bomgr_find_va(mgr, size, desc->alignment);
va.handle = bo->handle;
va.vm_id = 0;
va.offset = bo->va;
r = drmCommandWriteRead(rws->fd, DRM_RADEON_GEM_VA, &va, sizeof(va));
if (r && va.operation == RADEON_VA_RESULT_ERROR) {
- fprintf(stderr, "radeon: Failed to allocate a buffer:\n");
+ fprintf(stderr, "radeon: Failed to allocate virtual address for buffer:\n");
fprintf(stderr, "radeon: size : %d bytes\n", size);
fprintf(stderr, "radeon: alignment : %d bytes\n", desc->alignment);
fprintf(stderr, "radeon: domains : %d\n", args.initial_domain);
+ fprintf(stderr, "radeon: va : 0x%016llx\n", (unsigned long long)bo->va);
radeon_bo_destroy(&bo->base);
return NULL;
}
+ pipe_mutex_lock(mgr->bo_handles_mutex);
if (va.operation == RADEON_VA_RESULT_VA_EXIST) {
- radeon_bomgr_free_va(mgr, bo->va, bo->va_size);
- bo->va = va.offset;
- radeon_bomgr_force_va(mgr, bo->va, bo->va_size);
+ struct pb_buffer *b = &bo->base;
+ struct radeon_bo *old_bo =
+ util_hash_table_get(mgr->bo_vas, (void*)(uintptr_t)va.offset);
+
+ pipe_mutex_unlock(mgr->bo_handles_mutex);
+ pb_reference(&b, &old_bo->base);
+ return b;
}
+
+ util_hash_table_set(mgr->bo_vas, (void*)(uintptr_t)bo->va, bo);
+ pipe_mutex_unlock(mgr->bo_handles_mutex);
}
+ if (rdesc->initial_domains & RADEON_DOMAIN_VRAM)
+ rws->allocated_vram += align(size, 4096);
+ else if (rdesc->initial_domains & RADEON_DOMAIN_GTT)
+ rws->allocated_gtt += align(size, 4096);
+
return &bo->base;
}
static void radeon_bomgr_destroy(struct pb_manager *_mgr)
{
struct radeon_bomgr *mgr = radeon_bomgr(_mgr);
+ util_hash_table_destroy(mgr->bo_names);
util_hash_table_destroy(mgr->bo_handles);
+ util_hash_table_destroy(mgr->bo_vas);
pipe_mutex_destroy(mgr->bo_handles_mutex);
pipe_mutex_destroy(mgr->bo_va_mutex);
FREE(mgr);
mgr->base.is_buffer_busy = radeon_bomgr_is_buffer_busy;
mgr->rws = rws;
+ mgr->bo_names = util_hash_table_create(handle_hash, handle_compare);
mgr->bo_handles = util_hash_table_create(handle_hash, handle_compare);
+ mgr->bo_vas = util_hash_table_create(handle_hash, handle_compare);
pipe_mutex_init(mgr->bo_handles_mutex);
pipe_mutex_init(mgr->bo_va_mutex);
mgr->va = rws->info.r600_virtual_address;
- mgr->va_offset = rws->info.r600_va_start;
+ mgr->va_offset = rws->va_start;
list_inithead(&mgr->va_holes);
return &mgr->base;
unsigned *bankw, unsigned *bankh,
unsigned *tile_split,
unsigned *stencil_tile_split,
- unsigned *mtilea)
+ unsigned *mtilea,
+ bool *scanout)
{
struct radeon_bo *bo = get_radeon_bo(_buf);
struct drm_radeon_gem_set_tiling args;
*microtiled = RADEON_LAYOUT_LINEAR;
*macrotiled = RADEON_LAYOUT_LINEAR;
- if (args.tiling_flags & RADEON_BO_FLAGS_MICRO_TILE)
+ if (args.tiling_flags & RADEON_TILING_MICRO)
*microtiled = RADEON_LAYOUT_TILED;
+ else if (args.tiling_flags & RADEON_TILING_MICRO_SQUARE)
+ *microtiled = RADEON_LAYOUT_SQUARETILED;
- if (args.tiling_flags & RADEON_BO_FLAGS_MACRO_TILE)
+ if (args.tiling_flags & RADEON_TILING_MACRO)
*macrotiled = RADEON_LAYOUT_TILED;
if (bankw && tile_split && stencil_tile_split && mtilea && tile_split) {
*bankw = (args.tiling_flags >> RADEON_TILING_EG_BANKW_SHIFT) & RADEON_TILING_EG_BANKW_MASK;
*mtilea = (args.tiling_flags >> RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT) & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK;
*tile_split = eg_tile_split(*tile_split);
}
+ if (scanout)
+ *scanout = bo->rws->gen >= DRV_SI && !(args.tiling_flags & RADEON_TILING_R600_NO_SCANOUT);
}
static void radeon_bo_set_tiling(struct pb_buffer *_buf,
unsigned tile_split,
unsigned stencil_tile_split,
unsigned mtilea,
- uint32_t pitch)
+ uint32_t pitch,
+ bool scanout)
{
struct radeon_bo *bo = get_radeon_bo(_buf);
struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
/* Tiling determines how DRM treats the buffer data.
* We must flush CS when changing it if the buffer is referenced. */
if (cs && radeon_bo_is_referenced_by_cs(cs, bo)) {
- cs->flush_cs(cs->flush_data, 0);
+ cs->flush_cs(cs->flush_data, 0, NULL);
}
while (p_atomic_read(&bo->num_active_ioctls)) {
}
if (microtiled == RADEON_LAYOUT_TILED)
- args.tiling_flags |= RADEON_BO_FLAGS_MICRO_TILE;
+ args.tiling_flags |= RADEON_TILING_MICRO;
else if (microtiled == RADEON_LAYOUT_SQUARETILED)
- args.tiling_flags |= RADEON_BO_FLAGS_MICRO_TILE_SQUARE;
+ args.tiling_flags |= RADEON_TILING_MICRO_SQUARE;
if (macrotiled == RADEON_LAYOUT_TILED)
- args.tiling_flags |= RADEON_BO_FLAGS_MACRO_TILE;
+ args.tiling_flags |= RADEON_TILING_MACRO;
args.tiling_flags |= (bankw & RADEON_TILING_EG_BANKW_MASK) <<
RADEON_TILING_EG_BANKW_SHIFT;
args.tiling_flags |= (mtilea & RADEON_TILING_EG_MACRO_TILE_ASPECT_MASK) <<
RADEON_TILING_EG_MACRO_TILE_ASPECT_SHIFT;
+ if (bo->rws->gen >= DRV_SI && !scanout)
+ args.tiling_flags |= RADEON_TILING_R600_NO_SCANOUT;
+
args.handle = bo->handle;
args.pitch = pitch;
sizeof(args));
}
-static struct radeon_winsys_cs_handle *radeon_drm_get_cs_handle(
- struct pb_buffer *_buf)
+static struct radeon_winsys_cs_handle *radeon_drm_get_cs_handle(struct pb_buffer *_buf)
{
/* return radeon_bo. */
return (struct radeon_winsys_cs_handle*)get_radeon_bo(_buf);
unsigned size,
unsigned alignment,
boolean use_reusable_pool,
- enum radeon_bo_domain domain)
+ enum radeon_bo_domain domain,
+ enum radeon_bo_flag flags)
{
struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
+ struct radeon_bomgr *mgr = radeon_bomgr(ws->kman);
struct radeon_bo_desc desc;
struct pb_manager *provider;
struct pb_buffer *buffer;
memset(&desc, 0, sizeof(desc));
desc.base.alignment = alignment;
- /* Additional criteria for the cache manager. */
- desc.base.usage = domain;
+ /* Only set one usage bit each for domains and flags, or the cache manager
+ * might consider different sets of domains / flags compatible
+ */
+ if (domain == RADEON_DOMAIN_VRAM_GTT)
+ desc.base.usage = 1 << 2;
+ else
+ desc.base.usage = domain >> 1;
+ assert(flags < sizeof(desc.base.usage) * 8 - 3);
+ desc.base.usage |= 1 << (flags + 3);
+
desc.initial_domains = domain;
+ desc.flags = flags;
/* Assign a buffer manager. */
if (use_reusable_pool)
if (!buffer)
return NULL;
+ pipe_mutex_lock(mgr->bo_handles_mutex);
+ util_hash_table_set(mgr->bo_handles, (void*)(uintptr_t)get_radeon_bo(buffer)->handle, buffer);
+ pipe_mutex_unlock(mgr->bo_handles_mutex);
+
return (struct pb_buffer*)buffer;
}
struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
struct radeon_bo *bo;
struct radeon_bomgr *mgr = radeon_bomgr(ws->kman);
- struct drm_gem_open open_arg = {};
int r;
-
- memset(&open_arg, 0, sizeof(open_arg));
+ unsigned handle;
+ uint64_t size = 0;
/* We must maintain a list of pairs <handle, bo>, so that we always return
* the same BO for one particular handle. If we didn't do that and created
* The list of pairs is guarded by a mutex, of course. */
pipe_mutex_lock(mgr->bo_handles_mutex);
- /* First check if there already is an existing bo for the handle. */
- bo = util_hash_table_get(mgr->bo_handles, (void*)(uintptr_t)whandle->handle);
+ if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+ /* First check if there already is an existing bo for the handle. */
+ bo = util_hash_table_get(mgr->bo_names, (void*)(uintptr_t)whandle->handle);
+ } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ /* We must first get the GEM handle, as fds are unreliable keys */
+ r = drmPrimeFDToHandle(ws->fd, whandle->handle, &handle);
+ if (r)
+ goto fail;
+ bo = util_hash_table_get(mgr->bo_handles, (void*)(uintptr_t)handle);
+ } else {
+ /* Unknown handle type */
+ goto fail;
+ }
+
if (bo) {
/* Increase the refcount. */
struct pb_buffer *b = NULL;
goto fail;
}
- /* Open the BO. */
- open_arg.name = whandle->handle;
- if (drmIoctl(ws->fd, DRM_IOCTL_GEM_OPEN, &open_arg)) {
- FREE(bo);
- goto fail;
+ if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
+ struct drm_gem_open open_arg = {};
+ memset(&open_arg, 0, sizeof(open_arg));
+ /* Open the BO. */
+ open_arg.name = whandle->handle;
+ if (drmIoctl(ws->fd, DRM_IOCTL_GEM_OPEN, &open_arg)) {
+ FREE(bo);
+ goto fail;
+ }
+ handle = open_arg.handle;
+ size = open_arg.size;
+ bo->flink_name = whandle->handle;
+ } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ size = lseek(whandle->handle, 0, SEEK_END);
+ /*
+ * Could check errno to determine whether the kernel is new enough, but
+ * it doesn't really matter why this failed, just that it failed.
+ */
+ if (size == (off_t)-1) {
+ FREE(bo);
+ goto fail;
+ }
+ lseek(whandle->handle, 0, SEEK_SET);
}
- bo->handle = open_arg.handle;
- bo->name = whandle->handle;
+
+ bo->handle = handle;
/* Initialize it. */
pipe_reference_init(&bo->base.reference, 1);
bo->base.alignment = 0;
bo->base.usage = PB_USAGE_GPU_WRITE | PB_USAGE_GPU_READ;
- bo->base.size = open_arg.size;
+ bo->base.size = (unsigned) size;
bo->base.vtbl = &radeon_bo_vtbl;
bo->mgr = mgr;
bo->rws = mgr->rws;
bo->va = 0;
pipe_mutex_init(bo->map_mutex);
- util_hash_table_set(mgr->bo_handles, (void*)(uintptr_t)whandle->handle, bo);
+ if (bo->flink_name)
+ util_hash_table_set(mgr->bo_names, (void*)(uintptr_t)bo->flink_name, bo);
+
+ util_hash_table_set(mgr->bo_handles, (void*)(uintptr_t)bo->handle, bo);
done:
pipe_mutex_unlock(mgr->bo_handles_mutex);
if (mgr->va && !bo->va) {
struct drm_radeon_gem_va va;
- bo->va_size = ((bo->base.size + 4095) & ~4095);
- bo->va = radeon_bomgr_find_va(mgr, bo->va_size, 1 << 20);
+ bo->va = radeon_bomgr_find_va(mgr, bo->base.size, 1 << 20);
va.handle = bo->handle;
va.operation = RADEON_VA_MAP;
radeon_bo_destroy(&bo->base);
return NULL;
}
+ pipe_mutex_lock(mgr->bo_handles_mutex);
if (va.operation == RADEON_VA_RESULT_VA_EXIST) {
- radeon_bomgr_free_va(mgr, bo->va, bo->va_size);
- bo->va = va.offset;
- radeon_bomgr_force_va(mgr, bo->va, bo->va_size);
+ struct pb_buffer *b = &bo->base;
+ struct radeon_bo *old_bo =
+ util_hash_table_get(mgr->bo_vas, (void*)(uintptr_t)va.offset);
+
+ pipe_mutex_unlock(mgr->bo_handles_mutex);
+ pb_reference(&b, &old_bo->base);
+ return b;
}
+
+ util_hash_table_set(mgr->bo_vas, (void*)(uintptr_t)bo->va, bo);
+ pipe_mutex_unlock(mgr->bo_handles_mutex);
}
+ bo->initial_domain = radeon_bo_get_initial_domain((void*)bo);
+
+ if (bo->initial_domain & RADEON_DOMAIN_VRAM)
+ ws->allocated_vram += align(bo->base.size, 4096);
+ else if (bo->initial_domain & RADEON_DOMAIN_GTT)
+ ws->allocated_gtt += align(bo->base.size, 4096);
+
return (struct pb_buffer*)bo;
fail:
memset(&flink, 0, sizeof(flink));
if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) {
- if (!bo->flinked) {
+ if (!bo->flink_name) {
flink.handle = bo->handle;
if (ioctl(bo->rws->fd, DRM_IOCTL_GEM_FLINK, &flink)) {
return FALSE;
}
- bo->flinked = TRUE;
- bo->flink = flink.name;
+ bo->flink_name = flink.name;
+
+ pipe_mutex_lock(bo->mgr->bo_handles_mutex);
+ util_hash_table_set(bo->mgr->bo_names, (void*)(uintptr_t)bo->flink_name, bo);
+ pipe_mutex_unlock(bo->mgr->bo_handles_mutex);
}
- whandle->handle = bo->flink;
+ whandle->handle = bo->flink_name;
} else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
whandle->handle = bo->handle;
+ } else if (whandle->type == DRM_API_HANDLE_TYPE_FD) {
+ if (drmPrimeHandleToFD(bo->rws->fd, bo->handle, DRM_CLOEXEC, (int*)&whandle->handle))
+ return FALSE;
}
whandle->stride = stride;
ws->base.buffer_from_handle = radeon_winsys_bo_from_handle;
ws->base.buffer_get_handle = radeon_winsys_bo_get_handle;
ws->base.buffer_get_virtual_address = radeon_winsys_bo_va;
+ ws->base.buffer_get_initial_domain = radeon_bo_get_initial_domain;
}